← Engagements
04

AI Economics

"We get asked about AI at every board. I do not have an answer I would want to defend."

Fixed priceCEO or CFO

Context

Not the first wave
Fraud, risk and authorisation have run on machine learning in this sector for years, and that work generally earned its keep. The LLM wave is newer, larger, and much less examined.
Where most people are
For many fintech businesses the experience so far is one of two things. Either you approved a set of cases built on foundation models two or three years ago, and some are running while some quietly stopped. Or almost nothing got authorised beyond a few teams on Copilot. Either way, people across the business started using AI every day without anyone approving it.
In the numbers
A pilot gets signed off and runs its course. Getting the same thing working at scale is a different number: integration, engineering, and a cost per transaction nobody modelled. The plan and the actual have been compared. Rarely at a level of detail that changes anything.
The boundary
Holding the data is not the same as being allowed to use it for this. It sits across the acquirer, the PSP, the ledger and the CRM, and anything touching customer data brings in risk and compliance. In regulated financial services that is where rollout slows, long before anyone gets to the unit economics.

Approach

First

Find out where you actually are

Any reassessment has to start with an accurate picture of what is running, and in most businesses that picture does not exist in one place. What was approved, what actually got built, what is in daily use, and what people started using on their own without asking anyone. Coming at it from outside helps here, because the questions are easier to ask when you have nothing invested in the answers.

Then

Capex, running costs, revenue, strategic value

Four lines, taken in turn. What it cost to build. What it costs to run at the volumes you actually do rather than the volumes in the pilot. What revenue it has brought in, which is the line most businesses cannot answer with any confidence. And whether what you have built leaves you with a defensible and durable strategic position worth having, or just a bill.

Revenue attribution is usually the difficult one. Cost at scale is usually the surprise.

Next

What continues, what stops, and what waits

Each case gets a call rather than a score: continue, stop, restart on different assumptions, or leave it for now with a stated trigger for when to look at it again. The assumptions sit next to each call so that anyone can argue with the reasoning rather than the conclusion. In my experience that argument is worth more than the answer it produces.

Finally

The position, and how you argue for it

Cutting use cases is the easy half. The harder question is what your AI position actually is now, and you have to be able to put it to a board that has read the same headlines you have.

I draft versions of it with you: where the industry has moved since your plan was written, what that means for what you build and what you buy, and what you keep inside the business rather than hand to a model provider. Then the version that goes in front of the board, and a shareholder version if you need one. Where it has to be built out properly against a specific date, that becomes a Narrative Sprint.

I build as well as advise. Strafi is an AI-native product I have taken from nothing to production this year, so the inference bills, the data joins and the model choices in a review like this are ones I have dealt with myself.

Deliverables

The engagement produces two things. The first is a report: what is running, what each case has cost and what it has returned, and the call on each one. The second is a written position on AI, short enough to put in front of a board, with the numbers from the report standing behind it.

The financial model stays with you. Inference pricing and model capability both move quickly, and an answer that holds this year will not hold next, so the model is built to be re-run rather than read once.

  • A report on what is running, including what nobody approved
  • Capex, running costs, revenue and strategic value, case by case
  • A call on each case: continue, stop, restart on different assumptions, or wait
  • The financial model, yours to re-run
  • A written position on AI, drafted with you
  • A board version, and a shareholder version if you need one
  • A working session with the exec team

Scope

In scope: what is running, what it has cost and returned, the position it leaves you in, and how you argue for that position.

Out of scope: building the pipelines, technical audit of the models, selecting tooling, and writing acceptable-use policy. The build sits with your engineering or a vendor.

Terms

Fixed price, not a day rate. The scope is agreed up front and the price does not move with it. If the work takes longer than expected, that is my problem rather than yours, which is the right way round.

Quoted after a short call, once I understand what you are actually dealing with. That call produces a statement of work (SOW): the scope, the deliverables, the timing and the price. Nothing starts until we have both signed it. Invoiced on delivery. If the work is not what we agreed, I fix it before I invoice.

Timing is agreed on the call. It depends on scope, and I would rather set it against the date you are working to than quote a standard length. There is no discovery phase, so the first useful output comes early.

Most engagements end at delivery. Some clients keep me on a light retainer afterwards to keep the model current and to be available when the board asks something new. That is agreed at the end, not the start.

Where AI does part of the work, I say which part. Some of the analysis and model building uses AI tooling. The judgement, the method and the conclusions are mine, and I will tell you which is which if you ask.

Always open to a conversation.

martin@scalepointpartners.com
Or message me on LinkedIn.